Reproduce “Bi-level Score Matching for Learning Energy-based Latent Variable Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Fan Bao, Chongxuan Li, Kun Xu, Hang Su, Jun Zhu, Bo Zhang
The authors introduce BiSM for energy-based latent-variable models, where hidden variables make the data score intractable. It learns an approximate posterior and couples inference updates to model training, avoiding the partition function. On CIFAR10, BiMDSM with 20 latent dimensions reports FID-ES 26.62 ± 1.52, versus 30.19 ± 2.60 for reproduced MDSM; the comparison does not isolate latent variables’ effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.